Bring a real project.
Help test the method.

Yes. Applying is the first step toward a small beta trial, and it is one form, one send — the "Apply" button on the study page opens it: nine screening questions, the consent acknowledgments, your typed name, and the date. No code is needed to apply. Applying does not enroll you. The study screens your application, then you get either an enrollment confirmation email — which carries your code (it looks like ET-0906-03) — or a screened-out note. Then comes intake, which asks for your code. Do not begin a recorded session until you are confirmed and have completed intake — applying is step one, not the green light.

You must be 18 or older, be in the United States, have a real project of your own, and have access to a computer for the finishing steps. Working sessions can happen on a phone, tablet, or computer. If you take part, your name and your work's DOI will be published in the report. That is what makes the trial checkable by anyone. Your name is required either way; your work's DOI is required, and public, when you make a deposit.

The principal investigator (PI), William Stafford. The PI screens applications, sends every scripted message, closes enrollment, and rules on the kit. The PI may add additional principal investigators as the trial needs them. Each is named on the beta page and in the report the day they join, with the date, and is bound by every rule that binds the PI. The study's records are seen by the PI and any named co-PI, and by no other person. The study's mail is hosted by Proton, and the study's drafting uses AI tools under the standard, as the consent form says.

A real project you care about: a memoir, family history, cookbook, storytime book, creative work, research or nonfiction, school work, or a work project. The screening questions use fixed categories and a published rubric. One limit: a project that already has a recorded session under the protocol can't be the trial project — apply with a different one.

All four are pages on earnedtrust.org: consent.html, terms.html, duty-map.html, and privacy-note.html. Each shows its draft and date at the top, and each is linked beside the acknowledgment on the Apply form, so you can open and read every document you are asked to acknowledge before you acknowledge it.

The materials are free. There is no payment. Participants receive their own public, checkable project record and credit in the published report.

A light project is roughly 4–6 working hours across 4–6 weeks. A heavier project may take 15–50 hours across 8–12 weeks. Every project needs at least three recorded working sessions over at least two weeks. Your enrollment confirmation states your window as a number of days and a closing date. Hours are one cost of taking part; the other is the AI plan you'll be working on — see "What will I need?" just below.

What you'll need. An AI you already use, on a plan that can hold a long conversation — the protocol alone is about twelve thousand words, and you'll be working with it for weeks. Free plans often can't; a paid plan usually can, and a month or two usually covers a project — check your plan's cancellation terms. Check what your plan includes before you start. You'll also need a second AI service for one step — the hostile read — and a free account there is fine. Your writing AI will help you choose — Claude, ChatGPT, Gemini, Perplexity, Grok, Copilot, or whichever you already have; more than one critic is fine. If you truly have only one AI, a fresh conversation with it will do for that step; your record will say so; for work that makes checkable claims — a research piece, anything with sources — a second company is required, not optional. The finishing steps run two small scripts from the kit; that needs Python on your computer — a free, five-minute install your AI walks you through (see "I don't have Python"). Taking part may cost you a month or two of a subscription; the study pays nothing and asks for nothing.

Enrollment stays open until the principal investigator closes it. The PI does not close it before six people are enrolled, except by an announced early close — its date, the count, and the reason go on the beta page. Enrollment closes no later than twelve months from the first enrollment (from opening day, if nobody has enrolled by then). There is no fixed cap. The decision to close may not use any exit answer, DOI check, result, wall, or interim finding; no interim result is published before enrollment closes; the date, the count, and the reason are recorded before any analysis. If demand outruns the phase, the answer is a second phase. The phase ends when the last tester's window closes. Further phases are announced here.

A short tag the study makes by rule from the day your application arrived and its order that day — for example ET-0906-03. No code is needed to apply. If you are enrolled, your confirmation email carries your code, and every form after that asks for it first: intake, the exit form, Name and DOI correction, and withdraw. The form cannot check the code; the study can, and the check has two parts: the code must match one the study issued, and the form must come from the email address the code was sent to. A form that fails either is not filed, and answered with the notice: "NOTICE: your ETR code was not accepted. Please check your code and try again. The form must also come from the email address your code was sent to." The one reminder repeats the code. It is not an identity code and hides nothing: the name you give is the name that prints.

Sending the application authorizes only the handling of application data the consent form describes; consent to participate takes effect only on enrollment. If you are screened out, you are not a participant and none of the acknowledgments binds you. The study keeps your name, email, and exclusion reason until the trial closes, so it can recognise a repeat application; your form itself is deleted 30 days after the phase closes. Screened-out applicants print only as a count. A second application from a name on the screened-out list is answered with the screened-out note and nothing else. There is no list to join for a future phase; further phases are announced here.

It is welcome — memoirs and family books are exactly the projects this trial wants. That person's consent is your responsibility as the author, exactly as it is for any writer; the study does not ask for it, record it, or confirm it. The apply form and the exit form record only the fact that another living person's story is in the project, as a cohort count. Before you publish your record, run the privacy check with your AI and decide deliberately — a deposit is permanent.

No. Emails during the trial will not receive a response. The automatic reply points you here. Anything you send that the study did not ask for is not read and not answered. The five forms are different — see "Will I hear back after I send a form?" below.

If you do not complete the trial within your window, you will be logged as incomplete. That is a finding, not a failure. If you stop before the end and want your data in anyway, choose "I'm stopping here — count what I filed" on the exit form; your name and answers stay in, and you are counted as stopped. Stopped (with an exit form) and incomplete (without one) are each their own published count. Withdrawn is only your own act — the Withdraw form — counted separately.

No. During your trial, no human answers or guides you. Study messages are machine-sent or sent verbatim from published scripts. The trial is testing whether the public materials and your own AI are enough — an AI on the kind of plan "What will I need?" describes, above.

Your own AI, holding the Earned Trust Protocol, is your regular working helper — which is why it needs a plan that can hold a long conversation (see "What will I need?"). Ask it about the protocol and about the tools, accounts, devices, Zenodo steps, and your own project. You make every decision and every click.

We'd rather you didn't. The trial is testing whether the published materials and your own AI are enough, and help from another person is something we cannot measure — no one will know, and nothing is held against you, but it's the one thing that muddies the result. If you're stuck, tell your AI exactly what you're confused about, in your own words. Most of the time that's enough. When it isn't, that's a finding: mark it on the exit form's "where did you hit a wall?" table.

That is a study finding. On the exit form, mark this list in the fixed "where did you hit a wall?" table and select the nearest failure category. Then use your own AI and judgment to continue. Do not email a question; nobody will compose an answer. The question list is frozen for the phase from the day enrollment opens. Anything the trial teaches goes into the next phase's list, dated.

On your computer's own drive, or in a cloud folder set to always keep on this device. OneDrive and iCloud can "free up space" by turning files into placeholders; a script reading a placeholder gets nothing. Right-click the folder and choose Always keep on this device (OneDrive), or turn off Optimize Mac Storage (iCloud).

That is a sync conflict — the cloud saw two edits and kept both. Your INDEX (the file named 00 INDEX in your kit folder) is the record file: what the AI did, session by session. Keep the newer one, rename the other 00 INDEX — conflict copy <date>.md, do not delete it, and note it in your session log. The checksum list — your seal manifest — will show both files, and that's fine; it is honest.

Yes, for writing. But each draft you save into the kit's Drafts folder is a saved file with a version number, not a live document.

The kit includes small helper programs ("scripts"). Windows and Mac warn about anything downloaded; the scripts are plain text you can open and read before running — that is the point of them. Your AI can read them to you.

The kit needs it. Install it from python.org; on Windows, tick Add Python to PATH (a checkbox on the installer's first screen). Your AI walks you through it — this is outside the study, so your AI is who answers.

Nobody can recover it — not the study, not the kit, not the free locking program (7-Zip). Your unlocked folder is still on your computer, so you have lost nothing: lock a new box and deposit a new version.

Somewhere you will always find it. This is not your bank password — the box exists to keep your working record from strangers on the internet, not from your family. Write it on paper and put it with your important documents, or in the notes app you already use, or in your password manager. Keep it for as long as you keep the record (the standard's floor is five years) — a publisher may ask to see inside the box years from now. One rule: do not reuse a password you use anywhere else. And if you ever open the box for someone, you open it yourself and hand over the item they asked to see — never the passphrase.

For 30 days after publishing, yes — Zenodo lets you edit a record's files in that window without a new DOI, so a DOI alone does not prove the files never changed; the seal manifest does, and the study records the fingerprints it saw at each check. After the 30 days, corrections are a new version with its own DOI. Plan as if nothing can ever be taken back.

Apps compress images. Upload the original PNG file from your folder, never a forwarded copy.

No. Every email the study asks you for is written by a published form or a fixed button. There are five forms: apply (one form, one send), intake, the exit form (your DOI goes on it), Name and DOI correction, and withdraw — and every one after apply asks for your code. There is no blank email body and no field for prose. The first valid form of each kind is the one that counts. A later duplicate is logged in the contact ledger and cannot replace it.

Yes. Every form is answered by a scripted email: confirmed, or not accepted. Apply gets the enrollment confirmation with your code, or the screened-out note; intake gets the getting-started reply, sent by the PI once the intake's code and address check out; the exit form gets the scripted messages your outcome calls for — the confirmation of your exit (one opening if you completed the trial, another if you stopped), and, if your DOI check passes, your documented-participant letter — or, if the study's check finds your DOI or your name wrong, the correction email first; an exit that says you made no deposit, with the DOI field blank, gets the confirmation of your exit and nothing else; Name and DOI correction gets your letter or the "no deposit found" note; withdraw gets one note confirming it. If your code was not accepted — it did not match one the study issued, or the form came from an email address other than the one the code was sent to — you get the notice: "NOTICE: your ETR code was not accepted. Please check your code and try again. The form must also come from the email address your code was sent to." Nothing is composed: each answer is a script, and each is counted in the contact ledger — the study's published count of every message it sent or received during the trial.

No privacy is promised. Email is a postcard, not a locked safe. Send nothing you could not bear a stranger reading. The study's central commitment is that it never receives, holds, or reviews your work.

The study records listed in the consent form: your apply form (if you are screened out: your name, email, exclusion reason, and date on the screened-out list until the trial closes, and nothing else after 30 days), your code if you are enrolled, the signed enrollment documents, intake fields, the exit form (your DOI and your two lines — the public line, which prints, and the private line, which holds the seven report-private items and is kept, counted into totals, and never printed), any Name and DOI correction, and the dates of all of it. It never asks for your manuscript, transcripts, locked box, passphrase, or working files; it downloads your public deposit to run the five checks, and reads none of it. The study's records are seen by the PI and any named co-PI, and by no other person at the study; the providers we name handle what passes through them as any mail or AI service does. The study's mail is hosted by Proton, and the study's drafting uses AI tools under the standard, as the consent form says.

Never. No legitimate request asks for your passphrase, keys, whole archive, or project files. The study asks for nothing from inside your box, ever.

Treat it as exposed. Choose a new passphrase and make a new locked box from the working folder on your computer, by repeating the kit's locking step. A passphrase on an existing box cannot be changed; you make a new box. Do not send another message about it. Test the new box yourself. Keep the new passphrase only under your control, and stop using the old box for any future deposit. Your own AI, holding the protocol, can walk you through those published steps without receiving the passphrase. Mark it on the exit form: that is a finding about our materials, not a failure by you.

It is the set of files that helps your AI apply the protocol, make the QR record card and the seal manifest, and help you lock your working record into your locked box. Download the kit version named in your enrollment confirmation — the confirmation links that version's own file — and copy its checksum (the long string of letters and numbers labelled SHA-256) into your intake form. Use only that version, even if the site shows a newer one.

Not for you. Your trial runs on the kit version named in your enrollment confirmation, which links that version's own file. The kit may be updated during the phase when the PI rules an update is warranted; every version stays on the site at its own address, and the Reader verifies every version issued in the phase. Use the kit you downloaded. If the website shows a newer version during your trial, ignore it — your trial runs on the version named in your enrollment email. The kit is a file on your computer; nothing changes there when the website changes.

A few times over the whole trial — not every week, and not on a calendar. Here is the rhythm, so nothing surprises you. When you first sit down together, your writing AI asks what a "piece" is for your project: a chapter, a recipe with its story, one poem, or the whole thing. Then each piece gets one critique when you say it's finished — your writing AI writes a message carrying your draft and a request to be a tough critic; you copy it into a different AI; that AI writes back; you bring its reply home and your writing AI saves it word for word. If your work is long enough that the pieces have to hang together — a book rather than one story — the whole gets one more read at the end, and your AI tells you whether yours is that kind of project, so you never have to guess. A piece is read again only in one case: every critic agreed with you. That sounds like good news, but agreement is a warning sign, not a certificate, and it earns a second, tougher round. Each round takes about an afternoon, most of it reading. So, in numbers: once per finished piece, once more for the whole if it's long, again only if the critics were too kind — a few afternoons across the trial, and the rest of the time is yours.

Whenever you like. You're the author; your AI works for you, not the other way round. Say "I'd like this attacked" about anything — a paragraph, a chapter, an idea you're not sure of — and your writing AI will build the critic packet on the spot. Every critique you get is saved in your record, and the ones you ask for early are often the most useful, because you can still change your mind cheaply. The one thing that isn't optional is the required critique: each finished piece gets attacked once before you move on, and your AI will raise it when a piece is done. Asking for extra ones doesn't replace that; it just means you'll have more critics on record — which reads well, not badly.

The 00 INDEX is the record: what the AI did, session by session. It is what readers see, carried inside the QR card. The manifest is the seal: one line per file, a checksum (a fingerprint of the file's contents) beside each filename, no words. It proves nothing changed since sealing. The index says what happened; the manifest proves the files are unchanged.

Yes — and it is the easiest yes on this page. Sealing is not a ceremony; it is one command. Your AI runs it, and out comes your seal manifest: the fingerprint list, one line per file, a checksum beside each name, no words. That list is what turns a folder of files into a record a stranger can check. Without it, nobody — including us — can tell whether anything changed after you finished, and the study's deposit check simply stops. So every deposit carries it: your work, your QR record card, your seal manifest. The one thing that is optional is your locked box — whether you also publish a passphrase-locked copy of your working records on Zenodo. Your AI asks you that in plain words at the end, and either answer is a good answer. Seal: required in every deposit. Locked box: you make one and keep it at home — the exit form asks whether you opened it. Publishing it: your call. That is the whole rule.

Zenodo is the free public archive, run by CERN, where you create the deposit on your own account. Your deposit holds three files, and a fourth if you choose it: your work, the QR record card, the seal manifest — every deposit carries those three — and, if you chose to publish it, your locked box. The card's QR opens the AI-workflow index in the Reader. Nothing readable from your working record goes up.

A DOI is the permanent web address Zenodo gives your public deposit. It looks like 10.5281/zenodo.1234567. Of your project, it is the only thing you give the study — on your exit form. Zenodo gives every record two DOIs: one for the version you published and one for "all versions." The DOI you give must be the version DOI — the one shown on the record page for that version, not the "all versions" DOI.

No — it's optional. It's a nice thing to have, and it takes about two minutes to get one. Why authors have one: names collide and change. There are many John Smiths, and people marry, use initials, or publish under a pen name. An ORCID is a free, permanent number that belongs to one person, so every work they publish can be tied to them without ambiguity, across publishers and archives, for life. In this trial the name on your Zenodo record is the name that prints in the report; an ORCID beside it makes that name unmistakably yours. If you want one before your Zenodo deposit, go to orcid.org. Your AI should do a good job helping you fill out the form — ask it to walk you through the sign-up. Just never paste a password into the chat; the AI doesn't need it, and the record never should hold one. Zenodo has a box for the number on the author line; leave it empty if you don't have one.

Copy it from your Zenodo page — don't type it — and make sure it is the version DOI, not the "all versions" DOI. The exit form's recheck screen shows it back to you before you send. If it is still wrong when the study checks it, see "What if my DOI or my name is wrong?" below.

After your exit is in, the study opens the DOI you filed. Before the checks, the study confirms the record holds your Earned Trust deposit (your card and your manifest) and that the author name on the Zenodo record is the name on your exit line. If either does not match, you get one email saying which, and the correction form takes your name and DOI together. That email — the correction email — names what failed, one of four reasons: your DOI did not open · you gave the "all versions" DOI instead of the version DOI · the record does not hold your Earned Trust deposit · the author name on the record does not match the name on your exit line. It points you to the Name and DOI correction form, which asks for your code, your name, and your DOI; a correction is accepted until your window closes, or 14 days after that email, whichever is later. If the correction still fails, or no answer comes, one note tells you the record is counted as "no deposit found" — within completed or stopped, whichever the exit form said; your exit line still prints. Your report line and your documented-participant letter are generated from your filed or corrected exit line; nothing is substituted. Those are the only messages about a check.

If your exit form says you made no deposit and the DOI field is blank, your record is counted at once as "no deposit found" within completed or stopped, as your form says. You get the confirmation of your exit and nothing else — no correction email, no note, no documented-participant letter.

After your exit is in, it checks five public things, exactly as a stranger would: your DOI opens to your record; your files download; your QR card scans and opens the index in the Reader; your seal manifest is there and complete; and every fingerprint matches. The DOI you give must be the version DOI — the one shown on the record page for that version, not the "all versions" DOI. Before the checks, the study confirms the record holds your Earned Trust deposit (your card and your manifest) and that the author name on the Zenodo record is the name on your exit line. If either does not match, you get one email saying which, and the correction form takes your name and DOI together. The study does not judge your work or open your locked box. A named second reader re-runs the checks on a quarter of the deposits; both sets of results publish.

Your last step, and one form, one send. You fill it on the Reader page: fixed choices, dates, numbers, and your DOI. The page saves a draft in your browser as you type and warns you before you leave the page. When you press Submit, the page shows you back everything that will be used — your answers, your name exactly as it will print, your DOI — so you can check it before you send. Then one button opens a message already written, and Send is the only action left; if the button does not open your email program, the page shows the To, the Subject, and the whole Message, each with its own Copy button. When you press Submit, the exit form makes two lines from your answers. Your public line holds everything that prints — it goes in the report's appendix word for word, and it is the only thing in the QR image. Your private line holds the seven report-private answers; the study keeps it, counts it into totals, and never prints it. Both lines go in the same exit email. Your private line is held by the study and never printed; it is not encrypted, and email is a postcard. Compression is not encryption: anyone with the Reader can decode a line they hold. Keep your own copy.

The seven report-private items (your private line): the date you opened your locked box, the three questions about whether anyone asked for your passphrase, whether anything on your card makes you uneasy, where your sessions happened (phone, computer, or both), and whether another living person's story entered your work. They never print beside your name — only as totals. Your private line is held by the study and never printed; it is not encrypted, and email is a postcard. The form marks them.

Yes, at any time before publication, with no reason and no penalty. Use the published withdraw form; it asks for your code, and a filed withdrawal is confirmed by one scripted note. Withdrawal deletes what the study still holds, except the three signed documents and one ledger line. An exit form already sent is already raw data and still publishes; withdrawal cannot recall it — you are counted as withdrawn. A withdrawal filed after your exit cancels every letter not yet sent and keeps your filed exit line, which still prints. There is no way to stop the clock: your window is the one clock, and a trial not completed within it is logged as incomplete, not withdrawn. You receive the one scripted reminder, two weeks before your window closes, with your code and the exit link in it.

Treat it as permanent. The repository's rules may permit an owner-controlled removal window or narrow later requests, but requests can be refused, copies may already exist, and a citation tombstone remains. The study cannot remove a deposit made on your account.

If you take part, your name and your work's DOI will be published in the report. That is what makes the trial checkable by anyone. Your name is required either way; your work's DOI is required, and public, when you make a deposit. The name that prints is the author name on your Zenodo record — a pen name is fine, but it has to be the name on the DOI publication, because the DOI is the check. The name you enrolled with is held privately; your letter and citation come to you under it. The study does not verify whether the name on your record is legal, common, or chosen. It does check that the author name on the Zenodo record is the name on your exit line; if it is not, you get one email saying so, and the Name and DOI correction form takes your name and DOI together. You see exactly how you print on the exit form before you send, and that recheck is the last word — there is no later change window.

No. The exit form's recheck screen — the page that shows you back your name exactly as it will print and your DOI, under the line "Check this before you send. This is how you will appear in the report. Please make sure your name and your DOI are correct." — is the only correction point, and it comes before Send. That recheck is the last word. The one thing that can reopen it is the study's check: if the name on your Zenodo record does not match the name on your exit line, or your DOI fails, the correction email says which and the Name and DOI correction form takes your name and DOI together. After your exit you get the scripted messages your outcome calls for — the confirmation of your exit (one opening if you completed the trial, another if you stopped, and two parting prompts for your own AI), and, if your DOI check passes, your documented-participant letter — a receipt showing your citation line exactly as you confirmed it on the exit form's recheck screen, generated from your filed or corrected exit line, never retyped; nothing is substituted. The report's line for you is generated the same way. This is how you confirmed it on your exit form's recheck screen, and it is final. The letter says nothing about any check; results publish for everyone at once, in the report.

Yes. Failures and successes publish at equal weight, including failures of the study's own materials. That is the point of the trial.

There is no independent complaints route; this is a small study run by its principal investigator and any named co-PI, with no ethics board, and it says so. What you can correct is how you appear — on the exit form's recheck screen before you send, and only there, unless the study's check finds your name or your DOI does not match, in which case the Name and DOI correction form takes both. The checks are public; anyone can re-run them from your deposit.

Not yet. Enrollment opens when the design is published at its DOI (10.5281/zenodo.22547805), the website is locked in its final version, the first kit version is published with its checksum, the mail filter is tested, and we have finished our own test run — and this page will say so. You can apply now; applications wait and are screened once enrollment opens.